Lutz Bornmann has self-archived "Which Kind of Papers Has Higher or Lower Altmetric Counts? A Study Using Article-Level Metrics from PLOS and F1000Prime."
Here's an excerpt:
The present study investigates the usefulness of altmetrics for measuring the broader impact of research. Methods: This study is essentially based on a dataset with papers obtained from F1000. This dataset was augmented with altmetrics (such as Twitter counts) which were downloaded from the homepage of PLOS (the Public Library of Science). This study covers a total of 1,082 papers. Findings: The results from regression models indicate that Facebook and Twitter, but not Figshare or Mendeley, can provide indications of papers which are of interest to a broader circle of readers (and not only for the peers in a specialist area), and seem therefore be useful for societal impact measurement.